Work

Proto-value Functions: A Laplacian Framework for Learning Representation and Control in Markov Decision Processes